Intelligent monitoring and processing method and system for parallel computing storage blockage

By collecting and sorting events in parallel computing clusters, and combining congestion inference models and reinforcement learning strategy models, the risk of storage congestion in parallel computing is predicted and anti-resonance orchestration is performed. This solves the problem of storage resource congestion in multi-task parallel computing and improves the stability and efficiency of the computing process.

CN122240278APending Publication Date: 2026-06-19XIAN LOGGING DATA INFORMATION TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
XIAN LOGGING DATA INFORMATION TECHNOLOGY CO LTD
Filing Date
2026-05-13
Publication Date
2026-06-19

Smart Images

  • Figure CN122240278A_ABST
    Figure CN122240278A_ABST
Patent Text Reader

Abstract

This invention provides an intelligent monitoring and processing method and system for parallel computing storage congestion. The method includes: acquiring business write logs, system recovery logs, and storage status monitoring data in a parallel computing cluster to form a parallel write event sequence; extracting the time overlap, write bandwidth contention, metadata contention, cache write-back intensity, and queue backlog change between business write events and system recovery events to generate a congestion amplification time-series feature set; inputting the congestion amplification time-series feature set into a congestion inference model to determine the congestion amplification risk value and the target recovery event set; when the congestion amplification risk value reaches a preset processing condition, constructing a recovery scheduling state vector, using a reinforcement learning strategy model to output an action decision set, and performing anti-resonance orchestration processing on the target recovery event set; and generating node execution control instructions based on the anti-resonance orchestrated target recovery event set.
Need to check novelty before this filing date? Find Prior Art